Most patients who die from cancer do so because their cancer is resistant to available therapies, either intrinsically, or as it evolves in response to treatment. However, the fundamental mechanisms driving resistance remain largely unknown. Tumors are comprised of a complex multicellular ecosystem of malignant and non- malignant cells, and changes in their composition, states, spatial organization and interactions are central to therapeutic resistance. Thus, there is an enormous need to chart an atlas of a tumor's cells, their spatial organization and interactions as those change dynamically in resistance to therapy. Technological breakthroughs in spatial and single-cell genomics, including many innovations by our team, now put an atlas within reach, but harnessing this remarkable opportunity, requires collection of multiple spatial and single cell genomics data in clinical samples; novel study design strategies; new experimental and computational strategies to integrate across cellular and spatial data; algorithms to construct tumor atlases that capture the resistant state; and showing how to use an atlas to formulate and test new predictive models of resistance. The Boston Human Tumor Atlas Network Research Center (HTA-RC) will address each of these challenges by creating three comprehensive atlases of the cellular geography of human cancer to understand how changes in the tumor ecosystem lead to therapeutic resistance in: (1) Primary and acquired resistance to CDK4/6 inhibition in breast cancer; (2) Primary and acquired resistance to immune checkpoint blockade in metastatic melanoma; and (3) Primary resistance to immunotherapy in microsatellite stable (MSS) colorectal carcinoma (CRC) compared with microsatellite instable (MSI) CRC. All three tumors types tackle an unmet clinical need; have an approximately equal rate of resistance and response to allow comparisons between states; and harness significant clinical experience and build on substantial preliminary results at our center. To construct the atlases, we will collect at least 100 biospecimens per year from resections and biopsies of the three tumor types and analyze them with histopathological data, high-resolution spatial multiplex RNA and protein data, single- cell genomics data, and temporal clinical data. Our algorithms will recover key features of each data modality, and integrate them into a single atlas to determine what predicts and underlies resistance. We build on a well-established interdisciplinary team in two major cancer centers (DFCI, MGH) and four research institutions (Broad, Harvard, Stanford, Princeton). Our leadership (Haining, Regev) and Units comprise of foremost experts and pioneers in clinical genomics (Biospecimens; Johnson, Wagle), spatial and single cell genomics (Shalek, Rozenblatt-Rosen, Nolan, Zhuang), and computational biology and data science (Regev, Van Allen, Engelhardt). Our atlases will allow identification of predictive biomarkers of resistance in the tumor ecosystem, and therapeutic target discovery, targeting diverse facets of the complex tumor ecosystem.